T J  T  L U http://jtlu.

org
V. 14 N. 1 [2021] pp. 995–1017

Satisfaction with travel, ideal commuting, and accessibility to

employment
John P. Pritchard Karst Geurs Diego B. Tomasiello
Technion Israel Institute of University of Twente University of São Paulo
Technology k.t.geurs@utwente.nl diegobt86@usp.br
john.pr@campus.technion.
ac.il Adelaide Nardocci Prashant Kumar
University of São Paulo University of Surrey
Anne Dorothee Slovic nardocci@usp.br p.kumar@surrey.ac.uk
University of São Paulo
adslovic@usp.br Alex Hagen-Zanker
University of Surrey
Mariana Giannotti a.hagen-zanker@surrey.ac.uk
University of São Paulo
mariana.giannotti@usp.br

Abstract: This paper explores relationships between commuting times, Article history:
job accessibility, and commuting satisfaction based on a large-scale Received: July 8, 2020
survey applied in the Greater London Area (GLA), the municipality Received in revised form:
of São Paulo (MSP) and the Dutch Randstad (NLR). Potential February 1, 2021
accessibility to jobs is estimated under 3 different scenarios: reported Accepted: May 5, 2021
actual commuting times (ACT), ideal commuting times (ICT), and Available online: September 9,
maximum willingness to commute (MCT). In addition, binary logistic 2021
regression models, estimated using generalized linear modeling (GLM),
are performed to assess the impact of these temporal preferences on
the likelihood of being satisfied with commuting. As expected, ideal
and maximum commuting preferences strongly impact the volume and
spatial distribution of the measured accessibility to jobs. In the selected
case studies, estimated ICT-based job accessibility significantly decreases
total measured accessibility (60 to 100 percent), with those living in
the lowest accessibility zones impacted most. Furthermore, although
specific results varied between regions, the overall findings show an
association between ACT and satisfaction. Likewise, commuting mode
is found to be a strong predictor of travel satisfaction. Those actively
traveling in all three metropolitan regions tend to be more satisfied with
their commutes. Potential job accessibility is found to be only weakly
associated with travel satisfaction.

Keywords: São Paulo, London, Randstad, commuting satisfaction, ac-

cessibility, decay functions

Copyright 2021 John P. Pritchard, Karst Geurs, Diego B. Tomasiello, Anne Dorothee Slovic, Adelaide Nardocci, Prashant
Kumar, Mariana Giannotti & Alex Hagen-Zanker
http://dx.doi.org/10.5198/jtlu.2021.1835
ISSN: 1938-7849 | Licensed under the Creative Commons Attribution – Noncommercial License 4.0

The Journal of Transport and Land Use is the official journal of the World Society for Transport and Land Use (WSTLUR)
and is published and sponsored by the University of Minnesota Center for Transportation Studies.
996 JOURNAL OF TRANSPORT AND LAND USE 14.1

1 Introduction

In the past decade, the quality and quantity of accessibility research has rapidly increased. This is partly
due to the growing abundance of spatial and transport data that has enabled the measurement of tem-
porally disaggregated accessibility indicators. Chief among these emerging data sources is the growth of
general transit feed specification (GTFS) data for more accurate public transport (PT) travel times (for
a review, refer to Stępniak, Pritchard, Geurs, & Goliszek, 2019), and congestion data for more accurate
car travel times (e.g., Moya-Gómez & García-Palomares, 2017). In parallel, many studies have exam-
ined the relationships between commuting preferences, satisfaction with travel, and well-being (for an
overview, refer to Mokhtarian, 2019; Chatterjee et al., 2020). However, there have been few attempts to
link both research fields despite improved accessibility and improvements in travel satisfaction often be-
ing important goals for transport operators and policy makers. The work presented here aims to bridge
these two fields in order to provide additional insight into the subject at hand. This is achieved following
a two-pronged approach. First, the impact of preference-based travel times on the value of measured
potential job accessibility is explored. Then, the impact of job accessibility on commuting satisfaction is
analyzed, controlling for commuting preferences and other factors.
A key addition to the accessibility literature is that, relying on a unique data-set, zonal potential
accessibility is calculated under three different scenarios. Different decay functions are estimated for: (1)
current actual commuting times (ACT), (2) ideal commuting times (ICT), and (3) maximum willing-
ness to commute times (MCT). Recently, Chaloux et al. (2019) also recognized this gap in the literature,
and attempted to introduce satisfaction in accessibility measures. However, the methodology and analy-
sis presented here differs to that of Chaloux et al. (2019). There, accessibility was estimated considering
only the travel times of those who reported being satisfied with their commute. Here, participants were
directly asked what their ideal commuting times would be as well as their maximum potential thresh-
olds. This information is then used to generate three distinct scenarios. Furthermore, despite the body of
literature on commuting satisfaction, the impact of potential job accessibility on commuting satisfaction
has not been studied before. We hypothesize that a high level of job accessibility (by a specific mode) can
positively affect commuter satisfaction levels. This relates to the link between potential accessibility and
option values, as argued by van Wee and Geurs (2016). The option values concept indicates that people
in some cases value not just the actual use they make of the transport system, but also the availability of
options to travel or carry out activities.
By linking accessibility to employment, commuting desires, and commuting satisfaction this work
engages with an underlying assumption central to much of the accessibility literature: the general (im-
plicit or explicit) understanding that more accessibility is valuable and desirable. However, the value of
marginal increases in accessibility is not well understood. With the increase of case-specific studies that
measure geographic and modal accessibility differences, the question of how to understand and value
accessibility is beginning to attract more attention. It is hoped that further insights into this relationship
can be provided by simultaneously exploring zonal accessibility levels and the satisfaction of commuters
in three different contexts: the Greater London Area (GLA), the Randstad region of the Netherlands
(NLR), and the municipality of São Paulo (MSP).
The rest of this article is structured as follows. A brief overview of the literature is provided next.
This is followed by the description of the data and methods, with a particular focus on the accessibil-
ity and statistical analyses. The general model specification for the estimated binary logistic models for
commuting satisfaction is also presented. This is followed by the results for all three case study regions
(i.e., GLA, NLR, and MSP). First, reported commuting preferences (Section 4.1) and their impacts on
estimated potential accessibility (Section 4.2). Then, the associations between commuting satisfaction
and commuting preferences (Section 4.3). Lastly, the discussion and conclusions are presented.
Satisfaction with travel, ideal commuting, and accessibility to employment 997

2 Literature review

Within the travel behavior field, it is well established that travel times generally impact travel satisfac-
tion. However, there is growing evidence that commuting and subjective well-being should be studied
separately from general travel and well-being since they are likely to interact in distinct ways (Chatterjee
et al., 2020). More specifically, it has been found that longer commutes tend to result in lower satisfac-
tion in a variety of cultural contexts (e.g., Lancée, Veenhoven, & Burger, 2017; Zhu & Fan, 2018).
Furthermore, long travel times have been found to negatively impact mood (Morris & Guerra, 2015).
Despite this general perceived disutility of travel, it has also been long understood that travel has
some innate positive utility (e.g., Choo, Collantes, & Mokhtarian, 2005; Jain & Lyons, 2008). In fact,
most individuals are not interested in minimizing their travel time entirely (i.e., to zero), even while
commuting (Redmond & Mokhtarian, 2001). Furthermore, a range of factors including individual af-
finities, attitudes and preferences, as well as the transport mode have been shown to mitigate and impact
the negative relationship between trip duration and satisfaction (St-Louis et al., 2014). Modal choice is
particularly relevant on this front (Chatterjee et al., 2020). Active modes are generally more positively
related to travel satisfaction while PT tends to be negatively related. For example, in the Netherlands
(where one of the case studies is located), it has been found that individuals are often happy to commute
longer if they are either walking or cycling (Lancée et al., 2017). The impact of PT modes on satisfac-
tion is less straightforward. While typically there is lower satisfaction for PT users, train commuters have
been found to be significantly more satisfied than car drivers, bus users, and metro users (e.g., St-Louis et
al., 2014). On the other hand, long-duration bus commuting has been found to be deleterious for mood
(Morris & Guerra, 2015). Having said this, it has been argued that the modal impact may be overstated.
For example, De Vos (2019) argues that the level of satisfaction associated with specific modes is more
related to whether the preferred mode is used, than to the characteristics of the modes themselves.
Unsurprisingly, liking one’s commute mode has been found to be associated with higher commuting
satisfaction (Handy & Thigpen, 2019).
In terms of travel times, travel duration’s impact on commuting satisfaction is often mitigated by
a number of factors. Among these, the impact of self-selection and choices is particularly relevant. For
instance, Morris and Zhou (2018) could not find an association between longer commutes and life sat-
isfaction. The associations that they found for the longer commutes allude to why this may be the case.
They found that longer commuting patterns are associated with higher wages, suburban living, higher
home ownership rates, and marital status. All of these may be positively associated with subjective well-
being and satisfaction with life. These findings highlight that the context of the longer commute mat-
ters: the land use patterns, demographic patterns of suburbanization (i.e., urban vs. suburban or periph-
eral poverty), as well as the level of choice in these commutes. For example, in the Netherlands, longer
commutes are associated with higher income and higher educated individuals (e.g., van Ham, 2002). In
Belgium, De Vos and Witlox (2016) report that travel-liking individuals tend to live in areas that lend
themselves to longer commutes, while travel-adverse residents choose more accessible locations.
The interest in accessibility, commonly understood as the ease of reaching opportunities, contin-
ues to grow amongst researchers and policymakers. Within the field of location-based accessibility (cf.,
Geurs & van Wee, 2004), also referred to as primal accessibility (cf., Cui & Levinson, 2019), many mea-
sures have been developed. Here, we rely on the potential accessibility measure, introduced by Hansen
(1959), which weighs the available opportunities (e.g., jobs) using a trip decay function based on travel
time or cost. The decay function weighs the accessible opportunities relying on the estimated likelihood
that a trip of said length would be made. This function is usually estimated using trip diary data, and
is often incorrectly interpreted or assumed to represent commuting preferences. However, these trip
lengths, and the frequency of trips of certain lengths are unlikely to be primarily the result of prefer-
998 JOURNAL OF TRANSPORT AND LAND USE 14.1

ence. These lengths are a function of the land use distribution, and transport opportunities available.
For a large segment of the population the available alternatives can be accurately described as a restricted
choice set. Morris, Dumble, and Wigan (1979) already noted that a major disadvantage of using mea-
sures of actual travel behavior to evaluate the transport/land use system is the difficulty to disentangle the
influence of choices and constraints. This is readily apparent when comparing the accessibility provided
by different transport modes. The travel time decay of PT trips is often less steep than for car trips (or
other privately owned vehicles), and individuals with a preference or affinity for PT are more likely to
tolerate longer waiting times (Shaw, Malokin, Mokhtarian, & Circella, 2019). However it would be
asinine to assume that this is more likely to be the result of individual preferences for longer trips, and
not the result of the level of service provided by these systems. For this reason, it has been previously
argued that for cross modal comparisons, the same decay function should be used when comparing the
car and PT (e.g., Geurs & Ritsema van Eck, 2001; Pritchard, Tomasiello, Giannotti, & Geurs, 2019b).

3 Data and methods

3.1 Survey description

The data used in this analysis was collected between June 2017 and June 2018 in the MSP, the GLA,
and the NLR (which includes Rotterdam, Utrecht, The Hague and Amsterdam).
The survey explored residents’ perceptions of their built environment and the associated exposure
to air pollution in order to examine the relationship between job accessibility, air pollution exposure
and social justice. This was done by collecting information related to their choices, or lack thereof, in
selecting their residential and/or work locations. It was composed of three main sections: Commuting
and Employment, Housing, and Pollution and Health. The latter addressed perceptions of air pollution,
followed by sections on the respondents’ current employment and housing reality as well as their deci-
sion-making process for selecting their home and employment locations. General socio-demographic
information of the samples was also collected. The guiding principle during the design of the survey was
to collect a unique data-set that successfully bridged the gap between the existing secondary (land use
and transport) and experimental data (pollution measurements) in order to enhance and complement
the analysis. The analysis presented here relies on the commuting and employment section of the survey.
In terms of commuting, respondents in the GLA, NLR and MSP reported their commuting sat-
isfaction, their commuting duration preferences (ICT and MCT), and their ACT. As will be described
in more detail in Sections 3.3 and 3.4, this information is used to estimate potential accessibility models
and included as explanatory variables in the generalized linear modeling (GLM) estimations of com-
muting satisfaction in the three case study regions. Specifically, respondents answered the following
questions:

1. ACT: Roughly how long is your commute from home to work, in minutes, on a typical day?
2. ICT: Ideally, if you could choose, how long would you like your commute to be? (in minutes)
3. MCT: How long do you think you would be willing to commute? (in minutes)

Given the nature of the analysis, list-wise deletion was applied when working with the survey
data. Respondents that did not report their commuting preferences were not included in the statistical
analysis. In practice this meant that the unemployed sample was completely excluded (n=884). Ad-
ditionally, respondents who reported having a lower commuting willingness than their ideal commute
(i.e., MCT<ICT) were also eliminated, as were those reporting a longer ACT than their reported MCT.
Satisfaction with travel, ideal commuting, and accessibility to employment 999

It was assumed that these respondents had responded mistakenly, or misinterpreted the question. Fur-
thermore, for the statistical analysis, individuals who did not report their residential area, or satisfaction
were also excluded. The resulting sample size (n) after the deletions was 2 644 (nGLA = 716, nMSP = 579,
nNLR = 1227).

3.2 Sampling and case study areas

For the GLA and the NLR, the areas were stratified based on a two-stage stratification criterion fol-
lowing the general procedure presented in part by Cerin, Leslie, du Toit, Owen, & Frank (2007). The
stratification variables differed from the aforementioned project given the thematic differences. The two-
stage stratification consisted of ranking all the sub-areas of the GLA and the NLR in terms of income
and accessibility. Using these two criteria the areas were grouped into four income strata, each with two
substrata in terms of their accessibility levels.
The sampling in the MSP differed from the other case study areas since it involved in-person visits
to the surveyed households. Two-stage cluster sampling was performed with geographic (city zone) and
socioeconomic stratification of census tracts from the database of the 2010 IBGE Demographic Census.
The secondary units were the permanent households in each census tract. The census tracts were selected
by systematic sampling proportional to the number of households while households were selected by
simple random sampling. The city was divided into five zones and in four socioeconomic groups based
on the percentiles values of Municipal Human Development Index (MHDI) by census tracts obtained
by Ribeiro et al. (2019): Very Low ( < 10th ), Low 10th − 50th), High (50th − 90th) and Very High (>
90th of the MHDI). The survey was submitted and received approval from the Brazilian Ethical Com-
mittee, CAAE 62918316.0.0000.54211.

3.3 Accessibility analysis

Zonal potential accessibility to jobs is estimated following the general form proposed by Hansen (1959):

Ai = ∑ oj ∙f (tij)
j

Where Ai is the potential accessibility by a particular mode x in any location i; oj is the number of
opportunities (i.e., jobs) in any location j; and f(tij) is an impedance function based on travel time by the
selected mode x from location i to j.
Here, we estimate job accessibility by PT. GTFS is used to estimate travel time matrices for SP and
the NLR every 15 minutes during the morning peak (6:00 a.m. to 9:00 a.m.). Travel times are calculated
using the Network Analyst extension of ArcGIS and the network database is built using the Add GTFS
to a Network Dataset tool. For the GLA, WebCAT Time Mapping (TIM) is used because GTFS was
not available. Employment data-sets are taken from NS-SEC 2011 (ONS, 2017) in the GLA, LISA
2014 (Stichting LISA, 2014) in the NLR, and combined data from RAIS 2009 and CNEFE 2010
(García-López & Moreno-Monroy, 2016) is used in the MSP. For a detailed description of the method-
ology and data refer to Pritchard et al. (2019a).
The key addition to the accessibility literature is that, due to the use of a unique multi-national
survey, zonal potential accessibility is calculated under three different scenarios and in three different
contexts using the same survey questions. Different decay functions are estimated for: (1) ACT (2) ICT,
and (3) MCT using non-linear regression based on the preferences of the participants of the ASTRID
project survey. The functions are selected relying on the Akaike Information Criterion (AIC).
1000 JOURNAL OF TRANSPORT AND LAND USE 14.1

3.4 Statistical analysis

In addition to the geographical impacts in terms of the zonal accessibility, binary logistic regression
models, estimated using GLM, are performed in order to explore the relationship between commuting
time, commuting preferences, and personal characteristics. The modeling work was performed using the
R (version 3.6.1) statistical software (R Core Team, 2020) and the VGAM package (Yee, 2015).
Three models are presented, one for each of the case study areas, in order to understand how these
relationships vary between the different contexts. Commuter satisfaction is modeled as a binary variable.
The general hypotheses underlying the models are as follows:
Duration is expected to negatively associate with commuting satisfaction: All else equal, as
ACT increases, the likelihood of reporting being satisfied with commuting is expected to decrease.
Commuting preferences are expected to mediate this association: Those with higher tolerances,
i.e., a stated willingness to commute longer (MCT), are expected to be more likely to be satisfied with
their current commute. Additionally, those with longer ICT are also more likely to be more satisfied.
This is because they are more likely to have been able to select a combination of residential and employ-
ment locations that matches their preferences.
Additionally, the relationship between ACT and ICT is expected to be particularly relevant. Lower
dissonance between ACT and ICT has been found to be positively associated with commuting satisfac-
tion due to preferences being met (Humagain & Singleton, 2020; Ye, De Vos, & Ma, 2020).
Modal choice is expected to have an impact: In line with the literature (as described in Section
2), active travel is expected to have a strong positive relationship. A positive but less strong impact is
expected for individuals that report commuting with several modes, including active and transit, but
never commuting by private car.
Higher zonal accessibility is expected to positively impact satisfaction: This is expected to be a
complex relationship. Therefore, the level of accessibility is not expected to show a clear positive associa-
tion to commuting satisfaction. Instead, it is hypothesized that at certain levels, the accessibility is likely
to be linked to satisfaction, but not at the highest levels.
Reporting negative consequences will negatively associate with satisfaction: Those reporting
negative impacts due to commuting or other transport-related problems are likely to be less satisfied
with said commute. Although satisfaction can be influenced by expectations, it is hypothesized that
affirming that commuting is negatively impacting other aspects or your life, should result in lower likeli-
hood of still reporting being satisfied.
Higher subjective well-being is likely to positively associate with commuting satisfaction:
General satisfaction with life (i.e., subjective well-being) is likely to result in higher satisfaction with
commuting, but this is not expected to be particularly strong. The relationship between well-being and
travel satisfaction is complex, but it is hypothesized that those who are generally satisfied are more likely
to be satisfied with this or any other aspect of their lives.

3.4.1 Description of included target and explanatory variables

The target variable is a simple binary variable of commute satisfaction, where 1 is individuals who are
satisfied with their commute, and 0 those who are not. This variable was created from a single 5-point
Likert scale question. Respondents who reported being satisfied or very satisfied were assigned as “satis-
fied.” While the possibility of treating the variable as ordinal was explored, the response profiles in the
different case study regions necessitated the transformation of the variable. Particularly in the NLR, but
also in the GLA, there were very high numbers of individuals reporting being satisfied, and very few
being either very dissatisfied or dissatisfied. As a result, even with good model fit indices, the model was
Satisfaction with travel, ideal commuting, and accessibility to employment 1001

unlikely to correctly predict the individuals in particularly small groups. (More details and descriptive
analysis of these responses can be found in Section 4.3.1.)
Commuting times (ICT, ACT, and MCT) are included in 10-minute intervals, rounded to the
nearest tens, in order to more easily interpret the results. That is, an increase of 1 in the model represents
a 10-minute increase in commuting time. As can be seen in Table 6, when compared to the descriptive
results shown in Table 1, this leads to slight increases in the means, but not significantly so. Given the
benefits for the intelligibility of the results, as well as the fact that we tend to estimate time in intervals
(resulting in many numbers never being named as commuting times) it was decided that this was a
worthwhile transformation. Additionally, it should be noted that ACT and MCT were highly and
significantly correlated (0.78), while the correlation between ACT and ICT was less strong, albeit still
significant (0.59).
Additionally, in order to assess the dissonance between ACT and ICT, a binary variable is included
that differentiates respondents who are able to achieve their ICT (i.e., ACT ≤ ICT).
Modal choice is included as distinct binary variables: always commuting actively (i.e., walking and/
or cycling), always commuting by private car, mixed commuting that never includes the car (i.e., com-
muting by PT and other alternatives as well), and mixed commuting that may include the car. As high-
lighted in section 3.4, active commuters are expected to report higher satisfaction with their commutes.
As such, it was important to be able to distinguish this group in the sample. Likewise, it was relevant to
distinguish those who only commuted by car, since they were expected to perceive travel times differ-
ently and have different expectations that would impact their satisfaction. Therefore, those who reported
using multiple modes to regularly commute were divided to include a subset that, although not only
commuting actively, never used the car.
Zonal accessibility values (see section 4.2.2) are assigned to each of the respondents based on their
residential area. Although three accessibility values are estimated for each zone, they are highly corre-
lated amongst each other. Therefore, in order to avoid collinearity issues, only the potential accessibility
estimated based on ICT is included. Additionally, an analysis of the distribution of accessibility for the
respondents shows a right-skewed distribution with long tails. Given the long tails, the log of the ac-
cessibility is tested in the model. This is done in order to decrease the impact of very high accessibility
under the assumption that the relationship would not continue to increase as the values continue to
grow larger and larger. Coefficients on the natural-log scale are directly interpretable as approximate
proportional differences (Gelman & Hill, 2007).
In terms of transport difficulties, two different binary variables are included. The first is having ex-
perienced negative impacts due to commuting duration. In the survey, respondents were asked whether
they believed that the amount of time they spent commuting prevented or limited other aspects of their
life. We hypothesize that those who report these negative impacts to their lives are less likely to be satis-
fied with their commute. Additionally, car ownership is also included as a potential transport difficulty.
While car ownership can be a choice, it also limits commuting options.
To include a general measure of subjective well-being, the widely validated and often used Satisfac-
tion with Life Scale (SWLS) developed by Diener (1985) is used. The scores range from 5-35, with an
average score of 20-24 corresponding to average life satisfaction in developed countries. That is, they are
generally satisfied but would like some improvement at least in some areas (Diener, 2006). The results
can be used as either a continuous scale or as an ordinal scale, with particular ranges corresponding to
states of satisfaction. The ordinal scale is used in the models.
Finally, several personal characteristics are tested in the model. Age is modeled as a continuous vari-
able, and both the age and the squared age are tested. By including both it is then possible to model a
nonlinear relation of age, e.g., if there is positive effect of age and a negative effect of age squared then
this could be interpreted as a lessening of the effect as age increases. On the other hand, a positive effect
1002 JOURNAL OF TRANSPORT AND LAND USE 14.1

in both would indicate the effect grows stronger with age. Gender is included as a binary variable, with
women coded as 1 and men as 0. Educational attainment is also tested as a binary variable, to identify
respondents with more than a high-school education.

4 Results
4.1 Reported commuting preferences: ACT, ICT, and MCT

The descriptive results of the commuting duration preferences (ACT, ICT, and MCT) in each of the
case study regions can be found in Table 1. The NLR reported the lowest average ACT (≈ 24 min.), as
well as the lowest ICT (≈ 17 min.) and MCT (≈ 42 min.). The GLA and MSP reported similar ideal
commutes (≈ 20 min.). However, despite the respondents in the GLA reporting slightly longer average
current commutes (≈ 34 and 32 min., respectively), the Brazilians reported a higher maximum willing-
ness to commute(≈ 58 min.), than their counterparts in England (≈ 51 min.). The London results are
comparable to the National Travel Survey data for England, which indicates that the average one-way
commute duration is 31 min (DfT, 2018a) with 14% of commuters reporting commutes longer than
an hour (DfT, 2018b).

Table 1. Actual, ideal, and maximum commuting times (in minutes)

Type Mean Median Stdev

GLA Actual Commute Time (ACT) 34.3 min 30 min 23.7 min
Ideal Commute Time (ICT) 19.9 min 19 min 14.1 min
Max. Willingness to Commute Time (MCT) 51.1 min 50 min 27.9 min
MSP Actual Commute Time (ACT) 32.3 min 20 min 33.6 min
Ideal Commute Time (ICT) 20.4 min 15 min 16.9 min
Max. Willingness to Commute Time (MCT) 57.5 min 60 min 38.4 min
NLR Actual Commute Time (ACT) 23.7 min 20 min 16.2 min
Ideal Commute Time (ICT) 16.6 min 15 min 10.0 min
Max. Willingness to Commute Time (MCT) 41.5 min 40 min 21.2 min
Total Actual Commute Time (ACT) 28.7 min 25 min 24.0 min
Ideal Commute Time (ICT) 18.4 min 15 min 13.2 min
Max. Willingness to Commute Time (MCT) 48.0 min 45 min 28.8 min

On average, respondents reported current unidirectional times of approximately 29 minutes, but

a desire that was approximately 10 minutes shorter. They reported a maximum average tolerance of
almost 50 minutes (≈ 48 min.). Very few respondents reported an ICT of more than 30 minutes, with
most reporting an ICT between 10 and 20 minutes. In terms of MCT, respondents seemed to use
rougher estimates, with most responding with 30, 40, 45, or 60 minutes.
Furthermore, as is visible in Figure 1 and Figure 2 there is a general positive relationship between
both the ICT and MCT, and the current commuting patterns, i.e., as ACT increases so does the MCT
and the ICT. This points to individuals potentially anchoring their desires. Exploring the desired ideal
commuting in more detail (see Table 2), it becomes evident that most respondents do not currently
achieve their ICT. The percentage of the sample that reported a current ACT that was lower than their
ICT ranged between 5.7% in the GLA and 18.7%, in the MSP. This is in line with the findings of Ye
et al. (2020) in China, who reported 17.1%. The majority of the sample in each of the cases reported
Satisfaction with travel, ideal commuting, and accessibility to employment 1003

current commuting times higher than their ideal, but it was in the GLA where this was most marked
with ≈ 70% of the sample reporting commuting times higher than their ideal.

Figure 1. Scatter plot of desired ideal vs. current commuting

Figure 2. Scatter plot of willingness vs. current commuting

1004 JOURNAL OF TRANSPORT AND LAND USE 14.1

Table 2. Difference between actual and ideal commutes by mode (ACT vs. ICT)

Always Car Never Car Always Active Other Total

ACT < ICT GLA 4 (9.5%) 12(28.7%) 4(9.5%) 22(52.4%) 5.7%
MSP 8 (6.7%) 12(10.1%) 53(44.6%) 46(38.6%) 18.7%
NLR 40 (25.7%) 20(12.8%) 48(30.8%) 48(30.8%) 12.1%
ACT = ICT GLA 26(14.4%) 42(23.3%) 24(12.3%) 88(48.9%) 24.4%
MSP 49(26.4%) 43(23.1%) 75(40.3%) 19(8.9%) 29.3%
NLR 172(37.3%) 24(5.2%) 138(29.9%) 127(27.6%) 35.7%
ACT > ICT GLA 105(20.4%) 256(49.6%) 31(6.0%) 124(24.0%) 69.9%
MSP 80(24.2%) 187(56.7%) 29(8.8%) 34(10.3%) 52.0%
NLR 340(50.5%) 105(15.6%) 95(14.1%) 134(19.9%) 52.2%
Total 824(30.9%) 701(26.3%) 497(18.7%) 642(24.1%) n=2,664

Considering modal choice, in the NLR and the MSP, the largest segments of the population report-
ing a desire for longer commuting (i.e., ICT > ACT), were those who always commute actively (30.8%
and 44.6%). The smallest segment was made up of those reporting to always commute by car (or other
private mode), with the NLR and GLA reporting less than 10% and the MSP ≈ 26%. This is interesting
for two different reasons; first, despite the car providing higher accessibility to jobs in all three cities (e.g.,
Pritchard, Tomasiello, Giannotti, & Geurs, 2019a), and the car providing greater flexibility in terms of
departure times we see that most car commuters would still like to have shorter commutes. Secondly, the
higher satisfaction rates of active commuters is in line with previous findings; active commuters are often
happy to commute longer if they are either walking or cycling (e.g., Lancée et al., 2017).

4.2 ACT, ICT, and MCT-based potential accessibility

4.2.1 Travel decay functions based on commuting preferences

Travel decay functions were estimated relying on the ACT, ICT, and MCT. The results of the decay
estimations (measured in travel time) for each of the case study areas can be found in Table 3 and the
estimated curves in Figure 3.

Table 3. Estimated decay functions: ACT, ICT, and MCT

GLA MSP NLR

ICT Log-Logistic Negative Exponential Log-Logistic
α=-8.669 β=3.198 α=-0.060 α=-6.719 β=2.692
ACT Log-Logistic Negative Exponential Log-Logistic
α=-7.420 β=2.259 α =-0.034 α=-6.445 β=2.242
MCT Log-Logistic Log-Logistic Log-Logistic
α=-11.648 β=3.148 α=-9.032 β=2.460 α=-10.947 β=3.176
Satisfaction with travel, ideal commuting, and accessibility to employment 1005

Figure 3. Estimated variable decay functions in the GLA, MSP, and NLR

The estimated decay function for the current commuting (estimated using ACT) varies from the
estimation reported in other studies relying on large mobility travel surveys for the same areas (e.g.,
Pritchard et al., 2019a). This is to be expected, considering the differences in size of the sample and the
fact that they were collected in different years. However, the best fitting model is the same for each of
the case studies, albeit with differing values, i.e., the decay follows a similar pattern, but the degree at
which it decays varies. Given the focus of the analysis being presented (comparing against the reported
commuting preferences), all results and subsequent references to the estimation of the decay of current
commuting, will rely on the estimations found in Table 3. It is also interesting to note, that in the case
of the MSP, the MCT-based decay function, follows a similar pattern (log-logistic) to the other two case
studies, despite showing a negative exponential fit for the current ACT and the ICT.

4.2.2 Impacts of zonal accessibility measurement

These distinct decay functions result in vastly different accessibility profiles for each of the three sce-
narios. The total differences in accessibility are shown in Figure 4, and the relative changes in Figure 5.
As expected, the estimated ICT-based accessibility is significantly lower than the accessibility based on
the ACT, and the MCT-based accessibility is higher than the other scenarios.
When considering the ICT, on average, there was a zonal accessibility reduction of 85%, 73%, and
71%, respectively, in the GLA, MSP, and NLR. The centers of the cities are still found to have the high-
est accessibility to jobs, but under the ideal scenario the estimated values in these areas are comparable to
middling accessibility levels under the current ACT scenario in all three case areas.
As is evident from the descriptive statistics found in Table 4, the range of the zonal impact is
larger when considering MCT. In the MSP, despite respondents already having longer commutes, their
willingness to commute further was such that the associated average zonal increase in accessibility was
above 100%. As evidenced by Figure 5, this was due to the increases in the areas outside the CBD (and
the metro coverage). In the GLA, the opposite phenomenon is observed. There is a larger percentage
decrease when relying on ICT for the accessibility estimation than the equivalent percentage increase
when using the reported MCT. Finally, in the NLR, where respondents were closer to their desired
commuting time, we can see that the impact of using ICT to measure accessibility is much lower than
in the GLA and the MSP.
1006 JOURNAL OF TRANSPORT AND LAND USE 14.1

Figure 4. Geographical impacts of variable travel decay functions on accessibility measurements in the GLA, MSP, and NLR
Satisfaction with travel, ideal commuting, and accessibility to employment 1007

Figure 5. Percentage change in accessibility with variable decay functions in the GLA, MSP, and NLR

Table 4. Descriptive statistics of accessibility results

Mean Median Min Max Stdev

ICT-based GLA 147,867 86,170 4,877 833,223 162,249
MSP 185,659 142,679 160 856,921 161,185
NLR 24,222 18,317 108 139,857 21,014
ACT-based GLA 781,689 684,768 121,179 1,857,820 365,576
MSP 581,220 553,023 4,962 1,552,535 326,382
NLR 81,935 70,733 187 331,018 60,023
MCT-based GLA 1,231,619 1,112,586 113,142 2,664,795 569,624
MSP 1,156,596 1,120,045 73,987 2,609,393 326,382
NLR 149,231 125,332 233 601,484 114,510
1008 JOURNAL OF TRANSPORT AND LAND USE 14.1

As shown in Figure 6 and Table 4, the total range of accessibility values experienced by the NLR
sample is much more limited than in the MSP and GLA. Additionally, the relationship between the
accessibility results based on ACT and those based on MCT tend to follow a generally linear pattern in
all three cases. However, for those experiencing the highest levels of accessibility, there is a slight tapering
off in terms of the potential gains. This is particularly evident in the GLA. In terms of the ICT-based
accessibility, the opposite pattern is found. Although there are significant decreases in total measured
accessibility, it is those living in the highest accessibility zones who are impacted the least, reinforcing the
general spatial pattern found in Figure 4.

Figure 6. Accessibility impacts of commuting desires in the GLA, MSP, and NLR

Taken together, these findings show that currently, if individual commuting desires are considered,
very few areas of the case study regions actually provide their residents with high potential accessibility
to jobs. This is relevant because it calls into question how accessibility goals should be set within policy
making and how policy results should be analyzed.

4.3 Commuting satisfaction

4.3.1 Impacts of zonal accessibility measurement

As highlighted in Table 5, the level of reported satisfaction with commuting was particularly high in the
NLR, where 82% reported being at least satisfied, and 87% of those reporting that they were very satis-
fied with their commute. In the GLA, 71% reported being satisfied, with 62% of those in turn reporting
being very satisfied. On the other hand, in the MSP, less than half of the respondents (40%) reported
being satisfied with their commutes, with 52% of those reporting to be very satisfied.
Satisfaction with travel, ideal commuting, and accessibility to employment 1009

Table 5. Satisfaction with commuting: GLA, MSP, and NLR

Likert Scale GLA MSP NLR

Not Satisfied 1 13 (2%) 41 (6%) 10 (1%)
2 61 (8%) 47 (7%) 65 (5%)
3 118 (16%) 245 (39%) 119 (9%)
n not satisfied 192 (26%) 333 (52%) 194 (15%)
Satisfied 4 199 (27%) 121 (19%) 141 (11%)
5 326 (44%) 130 (20%) 924 (72%)
n not satisfied 525 (71%) 251(40%) 1065 (82%)
Missing n/a 21 (3%) 51 (8%) 32 (2%)
Total (n) 738 635 1291

Further exploring the modal choice, it is found that in the NLR 43% of the selected sample always
commutes by car, while these numbers are only 19% in the GLA and 23% in the MSP. In terms of
always commuting actively (either on foot or cycling), the lowest rates were found in the GLA (8%),
followed by the NLR (22%), and the MSP (27%). Amongst car commuters, those in the MSP were
the most dissatisfied. The majority of this subset (58%) reported being unsatisfied with their commute.
This was in contrast to the findings in both European regions, where the majority of car commuters were
found to be satisfied with their commutes (NLR: 79%; GLA: 71%).
Interestingly, when exploring the missing data in the samples, it is found that the great majority
of respondents who did not report their commuting satisfaction never commuted by car. In the GLA
none of the respondents who failed to report their commuting preferences commuted by private car. In
the NLR and MSP only 3 and 5 of the excluded individuals, respectively, reported commuting by car.
The link between satisfaction with commuting, and general satisfaction with life (i.e., subjective
well-being), was also explored. While there was a general positive trend between these two types of satis-
faction, the correlation was not particularly high (τ = 0.16 with the 5-point Likert commuting satisfac-
tion scale; τ = 0.13 with the binary commuting satisfaction).
In terms of gender, the descriptive statistics show that within the three samples there are not par-
ticularly large differences between the genders. In the NLR and the GLA women were slightly more
likely to report being satisfied with their commute when compared to their male counterparts (NLR:
84% vs. 78%; GLA: 73% vs.70%). In the MSP men were more satisfied than women, and the differ-
ences were slightly larger. Only 35% of women reported being satisfied with their commute, compared
to 44% men. Given these descriptive results it seems that gender is more likely to be a significant factor
in the MSP.

4.3.2 Binary logistic regression results: City specific estimations

The main results of the GLM estimations for the GLA, MSP, and NLR can be found in Table 7, Table
8 and Table 9, respectively. As explained in Section 3.4, the target variable in these models is commuting
satisfaction. Descriptive statistics of the sample can be found in Table 6.
Generally speaking, lower ACT is significantly associated with being satisfied with commuting.
As commuting time increases, the odds of being satisfied decrease. Furthermore, actively commuting is
positively associated with greater satisfaction. As expected, those reporting that their commuting nega-
tively impacted other areas of their lives are more likely to be unsatisfied. Finally, greater subjective well-
being is also associated with greater commuting satisfaction. The impact of accessibility and personal
1010 JOURNAL OF TRANSPORT AND LAND USE 14.1

demographics was much more limited. Commuting preferences (i.e., ICT and MCT) were found to
significantly impact commuting satisfaction in all of the case study regions, but the manner in which
this occurred varied.
Table 6. Descriptive statistics of the sample used in the GLM estimations (means)

GLA MSP NLR Total

Dependent variable
Satisfaction with commute (1/0) 0.71 0.40 0.82 0.69
Travel times (10 minute intervals)
Actual commuting time (ACT) 3.71 3.49 2.67 3.15
Ideal commuting time (ICT) 2.22 2.26 1.93 2.09
Maximum willingness to commute (MCT) 5.28 5.81 4.31 4.93
Modal choice
Always active (1/0) 0.08 0.25 0.22 0.19
Never car (mix) (1/0) 0.42 0.38 0.12 0.26
Always car (1/0) 0.18 0.21 0.43 0.31
Other combination (mix) (1/0) 0.32 0.16 0.24 0.24
Potential job accessibility (normalized)
ACT-based accessibility 0.40 0.27 0.26 0.30
ICT-based accessibility 0.07 0.07 0.08 0.07
MCT-based accessibility 0.64 0.56 0.46 0.54
Transport difficulties
Negative commuting impacts (1/0) 0.40 0.25 0.11 0.22
Car ownership (1/0) 0.77 0.60 0.90 0.79
Personal characteristics
Satisfaction with life scale (SWLS) 19.9 23.2 25.9 23.6
Woman (1/0) 0.53 0.50 0.53 0.52
Age 44.3 41.8 47.6 45.3
Education - More than high school (1/0) 0.49 0.25 0.53 0.45
Household size 2.64 3.33 2.64 2.81

In the NLR, having a larger maximum willingness to commute (i.e., MCT) was shown to increase
the odds of being satisfied (β = 0.17∗). Additionally, being able to achieve ICT had a very strong associa-
tion (β = 0.81∗∗) with commuting satisfaction, but the ICT, in of itself, was not significant. That is, hav-
ing a higher ICT did not make it more likely to report current commuting satisfaction. In the GLA, by
contrast, increases in the ICT (β = 0.27∗) were significantly associated with increased satisfaction. In the
MSP, the ICT and the MCT were not found to be significant, despite the direction of the relationships
being as expected. However, reporting a low dissonance between your ACT and ICT (i.e., achieving
your ICT) was strongly associated with satisfaction (β = 0.71∗∗). Additionally, the negative association
of increases in ACT was strongest in the NLR (βGLA = −0.41∗∗∗; βMSP = −0.25∗∗∗; βNLR = −0.65∗∗∗).
Satisfaction with travel, ideal commuting, and accessibility to employment 1011

Table 7. Associations with commute satisfaction (GLA)

Commute Satisfaction Coefficient a SE z-value Pr(>|z|)

Actual commuting time (ACT) -0.41*** 0.06 -6.74 0.001
Ideal commuting time (ICT) 0.27** 0.09 2.92 0.004
Always active (1/0) 3.84* 1.65 2.32 0.02
Log ICT-based accessibility 0.24• 0.14 1.71 0.09
Negative commuting impacts -1.07*** 0.20 -5.28 <0.001
Age 0.56* 0.23 2.47 0.01
Age2 -0.01* 0.005 -2.30 0.02
SWLS 0.14* 0.07 2.13 0.03
Interaction- ACT: Always active -1.17* 0.52 -2.23 0.03
Interaction- Age: Age2 0.00007* 0.00003 2.07 0.04
AIC 696

a
Note: *** p<0.001, ** p<0.01, * p<0.05, • p<0.1

Table 8. Associations with commute satisfaction (MSP)

Commute Satisfaction Coefficient SE z-value Pr(>|z|)

Actual commuting time (ACT) -0.25*** 0.07 -3.71 <0.001
ACT ≤ ICT 0.71** 0.24 3.05 0.002
Always active (1/0) 1.23* 0.57 2.14 0.03
Always car (1/0) -0.76* 0.35 -2.18 0.03
Never car (mix) (1/0) -0.84* 0.34 -2.50 0.01
Log ICT-based accessibility 0.18• 0.10 1.81 0.07
Negative commuting impacts -0.74** 0.28 -2.66 0.008
SWLS 0.25*** 0.07 3.81 <0.001
Interaction- ACT: Always active 0.68** 0.24 -2.77 0.006
AIC 601

Table 9. Associations with commute satisfaction (NLR)

Commute Satisfaction Coefficient SE z-value Pr(>|z|)

Actual commuting time (ACT) -0.65*** 0.10 -6.84 <0.001
Maximum willingness to commute 0.17* 0.08 2.25 0.02
(MCT)
ACT ≤ ICT 0.81** 0.26 3.16 0.002
Always active (1/0) 1.00* 0.43 2.32 0.02
Always car (1/0) -0.54** 0.20 -2.67 0.008
Negative commuting impacts -1.10*** 0.24 -4.61 <0.001
Age 0.02• 0.009 1.78 0.07
SWLS -0.25*** 0.06 4.08 <0.001
AIC 784
1012 JOURNAL OF TRANSPORT AND LAND USE 14.1

Taken together, these results seem to point to the different experienced realities and preferences in
the case study regions. There is a clear indication that the respondents in the Netherlands appear to be
much more averse to longer commuting times than their counterparts in London and São Paulo. In-
creasing commuting times is strongly negatively associated with satisfaction, and as discussed earlier, the
NLR is also the only case study region where higher MCT is associated with a greater likelihood of being
satisfied with commuting. This is in line with the descriptive results presented in Section 4.1. Respon-
dents in the NLR have shorter commutes than those in the GLA or MSP, prefer even shorter commutes,
and report lower tolerance for potential increases in their commute duration. In contrast, in the MSP,
respondents report a willingness to commute even further despite being generally dissatisfied with their
commute (only 40% report being satisfied), and having average commutes above 30 minutes (refer to
Table 6). It would appear that higher ACT is a less significant predictor on its own as a result of living in
a large mega city where there is a general expectation of long commuting times. In the GLA, among the
tested commuting preference variables (i.e., ICT, MCT, ACT ≤ ICT), having higher ICT was the only
significant predictor of satisfaction. This is in line with the accessibility results which showed that relying
on the ICT to estimate job accessibility significantly reduced the measured accessibility, while increases
in accessibility to account for MCT were less impactful (refer to Figure 5).
In terms of mode choice, and its impact on commuting satisfaction, a positive association of active
commuting was found in all three case studies, but with different intensities. The positive association
was lowest in the NLR (β = 1.00∗), followed by the MSP (β = 1.23∗) and then the GLA (β = 3.84∗),
where it was extremely strong. Furthermore, in the MSP and the GLA, the interaction between ACT
and active travel was also significant, i.e., commuting actively resulted in a diminishing impact of ACT
on satisfaction (βGLA = −1.17∗∗; βMSP = −0.68∗∗). In the NLR, although commuting actively was posi-
tively associated with satisfaction, it did not have a significant interaction with ACT. This is interesting,
and could be the result of the bicycle being considered a conventional travel mode to a greater extent
than in the other case study regions. While the cause for this is not further explored here, it is relevant to
note that despite active commuting mitigating the negative impacts of longer commutes on satisfaction
in the GLA and the MSP, this is unlikely to be related to the infrastructure or environmental quality. Cy-
cling infrastructure is generally of higher quality in the Netherlands, and it has been found that cyclists
in the MSP are more exposed to pollutants while commuting when compared to cyclists in the GLA
and Rotterdam, in the NLR (Brand et al., 2019).
Related to this, commuting by car was negatively associated with commute satisfaction in the NLR
(β = −0.54∗∗) and the MSP (β = −0.76∗). However, in the MSP, mixed commuting (i.e., respondents
who reported commuting by PT to varying degrees but who never used the car) had an even greater
negative association with satisfaction (β = −0.84∗). In the GLA, the impact of commuting by non-active
modes, be it car or PT, was not found to be significant.
In terms of subjective well-being, as expected, being generally satisfied with life is positively associ-
ated with higher odds of also being satisfied with commuting. However, the intensity of this effect was
not overwhelming. It was found to be highest in the NLR (β = 0.25∗∗∗), and the MSP (β = 0.25∗∗∗).
The lowest association was found in the GLA (β = 0.14∗). This lends further credence to the fact that
when assessing subjective well-being, satisfaction with commuting is unlikely to play a large role. How-
ever, as expected, those who reported that their commuting impacted other aspects of their lives were
much more likely to report being unsatisfied with commuting (βGLA = −1.07∗∗∗; βMSP = −0.74∗∗; βNLR
= −1.10∗∗∗). Similarly to the findings regarding the duration of the trips, this effect was strongest in the
NLR, followed by the GLA. The lowest impact was found in the MSP.
Finally, increases in the zonal potential accessibility to jobs were found to have limited impacts on
commuting satisfaction. Although no associations could be found in the NLR, the log of the ICT-based
accessibility was found to be positively associated to greater satisfaction in the GLA and the MSP (βGLA
Satisfaction with travel, ideal commuting, and accessibility to employment 1013

= 0.24•; βMSP = 0.18•). Similarly, the socio-demographic profiles of respondents did not result in par-
ticularly strong findings. Increases in the age were found to have a small positive effect in the NLR (β
= 0.02•). In the GLA both the age (β = −0.56∗), and the age-squared (β = −0.01∗∗) terms were found
to be significant, as well as their interaction (β = 0.000∗). This points to a small lessening of the effect
as the age increases.
No significant results were found in terms of gender, education, or other household variables such
as car ownership.

5 Discussion and conclusions

This paper examined relationships between commuting preferences, satisfaction with travel, general
well-being and potential job accessibility in three distinct case study regions: the Randstad region of the
Netherlands (NLR), the Greater London Area (GLA) and the municipality of São Paulo (MSP). Po-
tential job accessibility by PT was estimated using distance decay functions based on survey data where
respondents reported their actual (ACT) and ideal (ICT) commuting times, as well as their maximum
willingness to commute (MCT). This allowed us to disentangle the influence of preferences and con-
straints in the estimation of potential job accessibility.
In all three metropolitan regions, most respondents reported their ICT to be between 10 and 20
minutes, with only a few stating a desire to commute more than 30 minutes. This result is consistent
with the findings from Redmond and Mokhtarian (2001) for ideal commuting times in the United
States. The majority of the sample in each of the cases also reported current ACT exceeding their ICT.
This was highest in the GLA, where ≈ 70% were unable to achieve their ICT. This is particularly relevant
because ideal and maximum commuting preferences have a very large impact on the volume and spatial
distribution of accessible jobs. Job accessibility based on ideal commuting preferences significantly de-
creases total measured accessibility (60 to 100 percent), with those living in the lowest accessibility zones
being impacted the most. MCT based job accessibility, in turn, increases the measured accessibility (50
to 85 percent), in a generally linear pattern for all three cases. The significant reduction in accessibility
under the ICT-based scenario, and the high level of dissonance between ICT and ACT for a majority
of respondents, highlight that the current spatial-infrastructural constellation in these cities does not
facilitate commuting options in-line with the preferences of residents. In fact, the estimated accessibility
values for zones with the highest ICT-based accessibility to jobs, are comparable to middling accessibil-
ity levels under the ACT scenario in all three cases. Measuring ICT-based accessibility could therefore
provide policy makers with a way to set accessibility improvement targets, or benchmark the existing
accessibility of a region for different transport modes.
ICT and MCT were also found to impact the odds of being satisfied with commuting. However,
the way in which these preferences manifested themselves in each of the regions varied. The analysis of
commuting satisfaction reflects the transport, land use, and cultural differences between the GLA, MSP,
and NLR. However, similar results allow us to generalize some of the findings. In particular, a strong
positive association between commuting actively and commute satisfaction was found in all three case
study regions. Interestingly, this result was found to be lowest in the NLR, despite tending to have
higher quality active infrastructure (e.g., cycling lanes). It can be argued that a possible reason for this is
that the penetration of bicycles is such that it results in an increased likelihood of commuting by bike
despite this not being the particular preference of a respondent. In contrast, in the GLA and the MSP,
active commuters, and cycling commuters in particular, are more likely to be commuting in this manner
due to a strong preference for the mode. This would be in line with existing research where liking one’s
commute mode has been found to be associated with higher commuting satisfaction (Handy & Thig-
1014 JOURNAL OF TRANSPORT AND LAND USE 14.1

pen, 2019), and with the argument that the modal impact on commuting satisfaction has less to do with
the characteristics of the modes themselves, and is instead much more related to whether the preferred
mode is used (De Vos, 2019). This argument is further reinforced when considering that, although
increases in ACT were found to be negatively associated to satisfaction in all of the regions, this finding
was somewhat mitigated by actively commuting in the MSP and the GLA. This was not the case in the
NLR, where an interaction between ACT and active commuting was not found.
The impact of accessibility on commuting satisfaction was less than anticipated, despite the accessi-
bility of residential zones being important for residential location choices, and as a result, a driver of land
use and demographic changes in cities (e.g., Smith et al., 2020). Results show that increases in the zonal
potential accessibility to jobs have only limited impacts on satisfaction. The log of the desired accessibil-
ity was found to be only weakly positively associated to greater satisfaction in the GLA and the MSP,
while no impact could be found in the NLR. In addition to the models presented in the previous sec-
tions, models without the travel variables (i.e., ICT, ACT, MCT) were tested and these had very little ex-
planatory power and significance levels. This is a strong indication that in terms of reported satisfaction,
personal mobility is of much higher importance than zonal accessibility. In the literature, several papers
found associations between job accessibility and trip generation (e.g., Cordera, Coppola, Dell’Olio, &
Ibeas, 2016) and job accessibility and vehicle miles traveled (VMT) (e.g., Ewing & Cervero, 2010). In
this paper we find that the underlying psychology of travelers, expressed in satisfaction or dissatisfaction
with the commuting trip, is not strongly related to potential job accessibility, which effectively combines
the proximity to jobs and the travel time needed to reach those jobs. It is not surprising, however, that
individual factors are more indicative of personal satisfaction than zonal potential accessibility measures.
Having said this, further research is needed on this front.
Future research should be directed to further understand the relationship between accessibility,
travel satisfaction, and well-being. With the ever-growing list of studies measuring accessibility to dif-
ferent types of opportunities in different contexts, the question of how to understand and value acces-
sibility and (marginal) accessibility increases needs to become much more central in the literature. For
instance, the value of accessibility and of how it is associated to well-being and satisfaction may be better
understood through a sufficientarian lens. That is, that there is a “sufficient” level of accessibility to
guarantee continued well-being, and that the importance lies in being able to reach this level. Once this
threshold is reached the value of marginal increases in accessibility stop being as valuable to well-being,
and more specifically to satisfaction with travel. Likewise, increased accessibility that remains below a
“sufficient” level is also unlikely to highly impact satisfaction. More comprehensive job accessibility
measures could also be explored further. Commuting satisfaction is likely to be more strongly related to
the overall experienced daily travel impedance (including costs and comfort of traveling) rather than to
travel time alone, as was included in the accessibility analysis in this paper. Finally, we are lacking knowl-
edge on the underlying mechanisms of the ideal commuting patterns for different commuter groups
(e.g., high- and low-skilled workers) and the role of the spatial-infrastructure constellations as an enabler
or barrier for workers to achieve their commuting preferences.

Acknowledgements

This work was carried out under the framework of the ASTRID (Accessibility, Social justice and Trans-
port emission Impacts of transit-oriented Development) project that was jointly funded by NWO
(Netherlands; project number: 485-14-038), the ESRC (UK; grant number ES/N011481/1), and
FAPESP (Brazil; grant number: 2015/50128-9). JP Pritchard was a Zuckerman STEM Leadership
Research Fellow at Technion, and conducted the data collection and parts of the analysis as a researcher
at the University of Twente.
Satisfaction with travel, ideal commuting, and accessibility to employment 1015

References

Brand, V., Kumar, P., Damascena, A. S., Pritchard, J. P., Geurs, K., & Andrade, M. d. F. (2019). Impact
of route choice and period of the day on cyclists’ exposure to black carbon in London, Rotterdam and
São Paulo. Journal of Transport Geography, 76, 153–165. https://doi.10.1016/j.jtrangeo.2019.03.007
Cerin, E., Leslie, E., du Toit, L., Owen, N., & Frank, L. D. (2007). Destinations that matter: As-
sociations with walking for transport. Health & Place, 13(3), 713–724. https://doi.10.1016/
j.healthplace.2006.11.002
Chaloux, N., Boisjoly, G., Grisé, E., El-Geneidy, A., & Levinson, D. (2019). I only get some satisfac-
tion: Introducing satisfaction into measures of accessibility. Transportation Research Part F: Traffic
Psychology and Behavior, 62, 833–843. https://doi.10.1016/j.trf.2019.03.008
Chatterjee, K., Chng, S., Clark, B., Davis, A., De Vos, J., Ettema, D., … Reardon, L. (2020). Com-
muting and wellbeing: A critical overview of the literature with implications for policy and future
research. Transport Reviews, 40(1), 5–34. https://doi.10.1080/01441647.2019.1649317
Choo, S., Collantes, G. O., & Mokhtarian, P. (2005). Wanting to travel, more or less: Exploring the
determinants of the deficit and surfeit of personal travel. Transportation, 32(2), 135–164. https://
doi.10.1007/s11116-004-2219-8
Cordera, R., Coppola, P., Dell’Olio, L., & Ibeas, Á. (2016). Is accessibility relevant in trip generation?
Modelling the interaction between trip generation and accessibility taking into account spatial ef-
fects. Transportation, 44, 1577–1603. https://doi.10.1007/s11116-016-9715-5
Cui, M., & Levinson, D. (2019). Primal and dual access. Geographical Analysis, 52(3), 452–474. https://
doi.10.1111/gean.12220
De Vos, J. (2019). Satisfaction-induced travel behavior. Transportation Research Part F: Traffic Psychology
and Behavior, 63, 12–21. https://doi.10.1016/j.trf.2019.03.001
De Vos, J., & Witlox, F. (2016). Do people live in urban neighborhoods because they do not like to
travel? Analyzing an alternative residential self-selection hypothesis. Travel Behavior and Society, 4,
29–39. https://doi.10.1016/j.tbs.2015.12.002
DfT. (2018a). Average number of trips, miles and time spent travelling by trip purpose (Table nts0403).
London: Department for Transport. Retrieved from https://www.gov.uk/ government/statistical-
data-sets/nts04-purpose-of-trips
DfT. (2018b). Time taken to travel to work by region of workplace (Table tsgb0110). London: Department
for Transport. Retrieved from https://www.gov.uk/government/statistical -data-sets/tsgb01-modal-
comparisons
Diener, E. E. (1985, feb). The satisfaction with life scale. Journal of Personality Assessment, 49(1), 71–75.
https://doi.10.1207/s15327752jpa4901_13
Diener, E. E. (2006). Understanding scores on the satisfaction with life scale. Urbana-Champaign, IL:
University of Illinois.
Ewing, R., & Cervero, R. (2010). Travel and the built environment: A meta-analysis. Journal of the
American Planning Association, 76(3), 265–294. https://doi.10.1080/01944361003766766
García-López, M. Á., & Moreno-Monroy, A. I. (2016). Income segregation and urban spatial structure:
Evidence from Brazil. (Working paper No. 2016/08.) Caracas: CAF Development Bank of Latin
America.
Gelman, A., & Hill, J. (2007). Data analysis using regression and multilevel/hierarchical models. Cam-
bridge, UK: Cambridge University Press.
Geurs, K., & Ritsema van Eck, J. (2001). Accessibility measures: Review and applications (RIVM report
408505 006). Bilthoven, the Netherlands: RIVM - National Institute of Public Health and the
1016 JOURNAL OF TRANSPORT AND LAND USE 14.1

Environment.
Geurs, K., & van Wee, B. (2004). Accessibility evaluation of land-use and transport strategies: Review
and research directions. Journal of Transport Geography, 12(2), 127–140. http://doi.10.1016/j.jtran-
geo.2003.10.005
Handy, S., & Thigpen, C. (2019). Commute quality and its implications for commute satisfaction:
Exploring the role of mode, location, and other factors. Travel Behavior and Society, 16, 241–248.
https://doi. 10.1016/j.tbs.2018.03.001
Hansen, W. G. (1959). How accessibility shapes land use. Journal of the American Institute of Planners,
25(2), 73–76. https;//doi.10.1080/01944365908978307
Humagain, P., & Singleton, P. A. (2020). Investigating travel time satisfaction and actual versus ide-
al commute times: A path analysis approach. Journal of Transport & Health, 16, 100829. https//
doi.10.1016/j.jth.2020.100829
Jain, J., & Lyons, G. (2008). The gift of travel time. Journal of Transport Geography, 16(2), 81–89.
https://doi.10.1016/j.jtrangeo.2007.05.001
Lancée, S., Veenhoven, R., & Burger, M. (2017). Mood during commute in the Netherlands: What
way of travel feels best for what kind of people? Transportation Research Part A: Policy and Practice,
104, 195–208. https://doi.10.1016/j.tra.2017.04.025
Mokhtarian, P. (2019). Subjective well-being and travel: Retrospect and prospect. Transportation, 46,
493513. https://doi.10.1007/s11116-018-9935-y
Morris, E. A., & Guerra, E. (2015). Are we there yet? Trip duration and mood during travel. Trans-
portation Research Part F: Traffic Psychology and Behavior, 33, 38–47. https://doi.10.1016/
j.trf.2015.06.003
Morris, E. A., & Zhou, Y. (2018). Are long commutes short on benefits? Commute duration and vari-
ous manifestations of well-being. Travel Behavior and Society, 11, 101–110. https://doi.10.1016/j.
tbs.2018.02.001
Morris, J., Dumble, P., & Wigan, M. (1979). Accessibility indicators for transport planning. Transporta-
tion Research Part A: General, 13(2), 91–109. https://doi.10.1016/0191-2607(79)90012-8
Moya-Gómez, B., & García-Palomares, J. C. (2017). The impacts of congestion on automobile acces-
sibility. What happens in large European cities? Journal of Transport Geography, 62, 148–159. https://
doi.10.1016/j.jtrangeo.2017.05.014
ONS. (2017). The national statistics socio-economic classification (NS-SEC). Retrieved from https://
perma.cc/RL6N-H4CN
Pritchard, J. P., Tomasiello, D., Giannotti, M., & Geurs, K. (2019a). An international comparison of
equity in accessibility to jobs: London, São Paulo and the Randstad. Findings, February, 1–9. https://
doi.10.32866/7412
Pritchard, J. P., Tomasiello, D., Giannotti, M., & Geurs, K. (2019b). Potential impacts of bike-and-ride
on job accessibility and spatial equity in São Paulo, Brazil. Transportation Research Part A: Policy and
Practice, 121, 386–400. https://doi.10.1016/j.tra.2019.01.022
R Core Team. (2020). R: A language and environment for statistical computing [Computer software].
Vienna: The R Project for Statistical Computing. Retrieved from https://www.R-project.org/
Redmond, L., & Mokhtarian, P. (2001). The positive utility of the commute: Modeling ideal com-
mute time and relative desired commute amount. Transportation, 28(2), 179–205. https://
doi.10.1023/A:1010366321778
Ribeiro, A. G., Baquero, O. S., de Almeida, S. L., de Freitas, C. U., Cardoso, M. R. A., & Nardoc-
ci, A. C. (2019). Influence of vehicular traffic density on hospital admissions due to respiratory
tract cancer in the city of São Paulo, Brazil. Cadernos de Saúde Pública, 35(1), e00128518. https://
doi.10.1590/0102-311X00128518
Satisfaction with travel, ideal commuting, and accessibility to employment 1017

Shaw, F. A., Malokin, A., Mokhtarian, P., & Circella, G. (2019). Who doesn’t mind waiting? Examining
the relationships between waiting attitudes and person and travel related attributes. Transportation,
48, 395-429. https://doi.10.1007/s11116-019-10054-2
Smith, D., Shen, Y., Barros, J., Zhong, C., Batty, M., & Giannotti, M. (2020). A compact city for the
wealthy? Employment accessibility inequalities between occupational classes in the London metro-
politan region 2011. Journal of Transport Geography, 86(June), 102767. https://doi.10.1016/j.jtran-
geo.2020.102767
Stichting LISA. (2014). Landelijk InformatieSysteem van Arbeidsplaatsen (LISA). Retrieved from
https://www.lisa.nl/home
St-Louis, E., Manaugh, K., van Lierop, D., & El-Geneidy, A. (2014). The happy commuter: A com-
parison of commuter satisfaction across modes. Transportation Research Part F: Traffic Psychology and
Behavior, 26, 160–170. https://doi.10.1016/j.trf.2014.07.004
Stępniak, M., Pritchard, J. P., Geurs, K., & Goliszek, S. (2019). The impact of temporal resolution on
public transport accessibility measurement: Review and case study in Poland. Journal of Transport
Geography, 75, 8–24. htttps://doi.10.1016/j.jtrangeo.2019.01.007
van Ham, M. (2002). Urban form and job access: Disparate realities in the Randstad. In Job access,
workplace mobility, and occupational achievement (pp. 13–34). Utrecht, the Netherlands: Eburon
Publishers.
van Wee, B., & Geurs, K. (2016). The role of accessibility in urban and transport planning. In M.
Bliemer, C. Mulley, & C. Moutou (Eds.), Handbook on transport and urban planning in the developed
world (pp. 53–66). Cheltenham, UK: Edward Elgar Publishing. https://doi.10.4337/9781783471
393.00010
Ye, R., De Vos, J., & Ma, L. (2020). Analyzing the association of dissonance between actual and ideal
commute time and commute satisfaction. Transportation Research Part A: Policy and Practice, 132,
47–60. https://doi.10.1016/j.tra.2019.10.011
Yee, T. (2015). Vector generalized linear and additive models: With an implementation. New York:
Springer. https://doi.10.1007/978-1-4939-2818-7
Zhu, J., & Fan, Y. (2018). Commute happiness in Xi’an, China: Effects of commute mode, duration,
and frequency. Travel Behavior and Society, 11, 43–51. https://doi.10.1016/j.tbs.2018.01.001